Understanding AI cloud

The AI cloud, a concept only now starting to be implemented by enterprises, combines artificial intelligence (AI) with cloud computing. Two factors are driving it: AI tools and software delivering new, increased value to cloud computing which is no more just an economical option for data storage and computation but playing a significant role in AI adoption.

An AI cloud consists of a shared infrastructure for AI use cases, supporting numerous projects and AI workloads simultaneously, on cloud infrastructure at any given point in time. The AI cloud brings together AI hardware and software (including open source) to deliver AI software-as-a-service on hybrid cloud infrastructure, providing enterprises access to AI and enabling them to harness AI capabilities.

A significant amount of processing power is required to run AI algorithms, making it unaffordable for many enterprises, but this deterrent is being eliminated by the recent availability of AI software-as-a-service, on the lines of software-as-a-service or infrastructure-as-a-service.

Why AI cloud

The most compelling advantages of AI cloud are the challenges it addresses. It democratises AI, making it more accessible. By lowering adoption costs and facilitating co-creation and innovation , it drives AI-powered transformation for enterprises.

The cloud is veritably becoming a force multiplier for AI, making AI-driven insights available for everyone. Besides, though cloud computing technology now is far more prevalent than the use of AI itself, we can safely assume that AI will make cloud computing significantly more effective.

AI-driven initiatives, providing strategic inputs for decision-making, are backed by the cloud’s flexibility, agility, and scale to power such intelligence massively. The cloud dramatically increases the scope and sphere of influence of AI, beginning with the user enterprise itself and then in the larger marketplace. In fact, AI and the cloud will feed off each other, aiding the true potential of AI flower through the cloud.

The pace of this will depend only on the AI expertise that enterprises can bring to bear in their workplace activities, for the cloud is already here and seeping everywhere. Investments enterprises make in using AI will gain multi-fold returns through the cloud; this makes the AI cloud very alluring.

Inherently AI workloads are computing and memory intensive, be it training new models or running existing models. Workloads for video, speech or large text data need huge memory and processor footprint that can be easily provisioned with cloud scaling resources in an automated way. Clients can benefit from these AI services, solutions with access to curated datasets, trained models, and an end-to-end tool stack.

A cloud-hosted AI platform has multiple layers, the bottom-most being the infrastructure management layer, critical for ensuring that computing is cloud and hyperscaler-agnostic and scalable on-demand.

Next comes the engineering lifecycle management layer, key in making AI vendor and technology workbench agnostic, driving standardisation and de-skilled deployment. It ensures optimised hardware use and that deployment is agnostic regardless of processor (CPU/GPU) architecture.

The middle layer governs AI and the digital workforce responsibly while providing operational visibility.

Then comes the API layer, allowing the larger developer community to use pre-defined base models, thereby ensuring standardisation or ‘uberising’ technology services on demand.

The topmost layer is the experience layer that allows access to assets, enablement, and expertise, facilitating collaboration, re-use, learning, and crowd-sourcing.

Future-proofing the AI cloud

Organisations need to build an enterprise-grade AI platform strategy with a software stack bringing together multiple technologies and stitching things together in a systemic manner to scale AI adoption, and crowd-source development to break silos by de-skilling.

To be future-ready, organisations need to have an approach that allows them to be agnostic to critical elements such as infrastructure, be it from hyperscalers or open source providers for models, algorithms, and AI tooling stacks. They need to standardise the management of models, datasets, and data pipes at the enterprise level. This, so that a switch can be made without hindering the use of business applications because of a change in any of the underlying layer components discussed earlier.

Enterprise software integrated with AI is now the primary way of using AI and such software is increasingly cloud-based, helping in making AI cloud more real. Future exists in collaborating with enterprises to create domain specific scenarios and models for different industries such as telecom, manufacturing, healthcare, finance, and insurance. Verticals that can help in quickly weaving AI capabilities to realise their vision of becoming an AI-first enterprise.

The writer is Senior Vice President, Service Offering Head - Energy, Communications and Services, AI and Automation, Infosys